Machine learning models demonstrated moderate discriminative ability for predicting postoperative atrial fibrillation after CABG, with a pooled AUC of 0.84 (95% CI 0.80-0.87).
Meta-Analysis
Do machine learning models accurately predict postoperative atrial fibrillation in patients following coronary artery bypass grafting?
Machine learning models show moderate accuracy for predicting postoperative atrial fibrillation after CABG, but routine clinical adoption is currently limited by methodological shortcomings and a lack of robust external validation.
Effect estimate: AUC 0.84 (95% CI 0.80-0.87)
Postoperative atrial fibrillation (POAF) is a common complication following coronary artery bypass grafting (CABG) and is associated with adverse clinical outcomes. Traditional risk prediction models show limited accuracy, prompting increasing interest in machine learning based approaches. This systematic review and meta analysis aimed to evaluate the diagnostic performance, methodological quality, and clinical applicability of machine learning models for predicting POAF after CABG. A comprehensive literature search identified observational studies developing or validating machine learning or advanced statistical models for POAF prediction after CABG. Diagnostic performance measures were pooled using random effects bivariate models. Risk of bias and applicability were assessed using PROBAST and PROBAST AI. Between study heterogeneity, sensitivity analyses, meta regression, and publication bias were evaluated. Twelve studies were included in the quantitative synthesis. The pooled sensitivity was 0.73 (95% CI 0.64–0.80) and pooled specificity was 0.83 (95% CI 0.73–0.90). The pooled area under the receiver operating characteristic curve was 0.84 (95% CI 0.80–0.87), indicating moderate discriminative ability. Substantial heterogeneity was observed (generalized I 2 = 87.3%). Most studies were judged at high risk of bias, primarily due to limitations in analysis methods and validation strategies, resulting in overall certainty of evidence rated as moderate to low. Machine learning models demonstrate moderate accuracy for predicting POAF after CABG but are limited by heterogeneity, methodological shortcomings, and restricted external validation. Further rigorously designed and prospectively validated studies are needed to support clinical implementation. • Machine learning models show moderate accuracy for predicting postoperative atrial fibrillation after CABG. • Substantial heterogeneity and high risk of bias limit confidence in current evidence. • Robust external validation is required before routine clinical adoption.
Eini et al. (Sat,) conducted a meta-analysis in Postoperative atrial fibrillation (POAF) following coronary artery bypass grafting (CABG). Machine learning and advanced statistical models was evaluated on Diagnostic performance (area under the receiver operating characteristic curve) for predicting POAF (AUC 0.84, 95% CI 0.80-0.87). Machine learning models demonstrated moderate discriminative ability for predicting postoperative atrial fibrillation after CABG, with a pooled AUC of 0.84 (95% CI 0.80-0.87).